• Aucun résultat trouvé

In the end, do choices truly affect preferences?

4. GENERAL DISCUSSION AND CONCLUSIONS

4.3. Perspectives and limitations

4.3.1. In the end, do choices truly affect preferences?

We have already mentioned at several occasions the importance of the issue raised by Chen and Risen (2010) i.e., that measured postchoice preference change may not necessarily be indicative of an influence of choice on preferences. Currently, this issue is still under active investigation.

This issue might even not be restricted to the field of preference modulation induced by choice. In a different literature context, regarding the distinction between liking/wanting (in very few words, liking refers to the hedonic impact of a stimulus or a situation while wanting refers to the incentive salience of this stimulus or this situation;

e.g. Berridge, Robinson, & Aldrige, 2009), some authors have brought arguments linked with Chen and Risen’s. Dai et al. (2010) thus considered that “some previously collected data that can be interpreted in terms of the wanting-liking distinction can also be interpreted in terms of the liking measure being too insensitive. For instance, you may think that you would get equally pleasure from eating tiramisu or zabaglione right now, and rate them as equally pleasurable. When forced to choose one or the other, a tiny preference for one that did not show up on the rating may tip your decision. This could reflect a wanting-liking divergence, or it could just reflect a difference in measurement sensitivity” (p. 332).

What do our empirical results suggest regarding Chen and Risen’s (2010) argument that classically reported rating modulation in the free-choice paradigm might be observed without any preference modulation ? First, let us note that this issue directly affects Experiments 1 and 5. Data collection for those experiments occured before the release of Chen and Risen (2010)’s paper, hence no control for this issue was run.

Second, let us emphasize that Chen and Risen (2010)’s criticism is supported by our experimental results. More precisely, if we considered how each of our experiments controls for Chen and Risen (2010)’s criticisms, here is what we can conclude.

In Experiment 2, we investigated the impact of blind choices, i.e. choices participants had the feeling to make if they were not, on olfactory preferences. This methodology is interesting because it allows controlling for the criticisms of Chen and Risen (2010), as « choices » are dissociated from pre-existing preferences. Our results failed to demonstrate significant postchoice preference modulation, unlike other experiments in the literature (Egan et al., 2010 ; Sharot et al., 2010).

In Experiments 3-4, our results suggest that the free-choice paradigm can be used to measure a true postchoice preference modulation if the appropriate controls are added.

In these two experiments, we compared postchoice (=second rating) preference modulation in the regular Rating-Choice-Rating (RCR) sequence to the modulation of second ratings in the RRC sequence (where choice can reveal but not affect preferences). This modulation was higher in the RCR condition than in the RRC

condition for rejected odors, showing that choices had an effect that went beyond simply revealing preferences – it affected them. Note however that this modulation was much less strong than what was believed before Chen and Risen (2010) made their methodological argument (see Holden, in press). Moreover, post-choice pleasantess rating modulation was significant in both the RCR ans the RRC sequences, again emphasing the importance of Chen and Risen’s (2010) point.

In Experiment 6, we used three sequences of measurement – Rating-Choice-Rating-Rating (RCRR), Rating-Choice-Choice-Rating-Rating-Choice-Rating-Choice-Rating-Rating (RRCR) and Rating-Choice-Choice-Choice-Rating-Rating- Rating-Rating-Rating-Choice (RRRC). For chosen odors, preference increase between Rating 1 and Rating 2 in the RCRR sequence (where a choice phase has taken place in between) tended to be higher than the one observed in RRCR and RRRC (where no choice had occured in between), suggesting that choice does impact preferences. However, similarly to Experiments 3 and 4, pleasantness ratings modulations between Rating 1 and Rating 2 were significant in all sequences of measurement, again emphasing the importance of considering in Chen and Risen’s (2010) point when investigating preference modulation induced by decision-making processes.

In Experiment 9, we set up a design where choice could not possibly reflect participants’ preferences for the odors presented. In this context, choice did have an impact on subsequent preferences.

Taken together, the observation of post-choice pleasantness rating modulation in the RRC, RRCR and RRRC conditions, where choice cannot affect preferences, illustrates the importance of Chen and Risen’ s (2010) contribution. Our results however suggest that choice can have a true impact on preferences. Despite the controversy raised by Chen and Risen (2010), we demonstrated that the modulation of pleasantness ratings is higher when following a difficult choice than when simply recording repeated ratings, and then separating trials according to a choice made thereafter (Experiments 3-4). Moreover, the impact of choice on preferences was also demonstrated by using a new design, derived from the effort-justification paradigm (Aronson & Mills, 1959). In Experiment 9, participants were making choices about odors before smelling them. In this context, choice cannot be a reflection of pre-existing preferences. Yet, postchoice preference modulation was still reliably shown and demonstrated to be stable over time. The observed revaluation process in the free-choice paradigm may consequently be made up of two processes: a spreading of alternatives

based on the problem identified by Chen and Risen (2010), but crucially, also of a spreading of alternatives linked to preference modulation induced by choice.

Current psychological studies on preference modulation induced by choice are now undergoing tranformations to deal with this issue, i.e. pre-existing preferences could explain post-choice preference modulation, trying to adopt a reliable and nonambiguous strategy to study preference modulation induced by choice. For instance, in a working paper, Arad investigated whether choice can increase the likelihood of a chosen stimulus to be chosen again in the future, with new alternatives, while controlling for the issue raised by Chen and Risen (2010). As already mentioned in the introduction of this work (section 2.3.2), the choice blindness paradigm has also recently been modified to avoid Chen and Risen’s (2010) criticisms (Johansson, Hall, & Chater, 2011). The question of which paradigms can most elegantly demonstrate an impact of preferences is still currently a burning issue (e.g., Holden, in press).

Chen and Risen’s mathematical proof does not however seem to have unanimous support among economists. For instance, Alós-Ferrer and Shi (2012) exposed in their abstract that “the mathematical proof was incorrect”. They furthermore explained that in constructing “a formal example of a choice model with mistakes fulfilling all assumptions in Chen and Risen (2010)”, ratings spreading does not follow. This means that “the rating spread observed in the experiments with the free-choice paradigm may be an informative signal of cognitive dissonance reduction”. A mathematical debate seems emerging with this manuscript’s release.

Thus, taken together with other results (Egan et al., 2010; Sharot et al., 2010), our results argue for the importance of Chen and Risen’s (2010) contribution. Despite this controversy, our results have made the case that choice impacts preferences. This demonstration goes beyond a given paradigm, as it was demonstrated both by adding a control sequence to the classical sequence in the free-choice paradigm, and also by using a new paradigm not subject to Chen and Risen’s (2010) criticism.

Next, we will discuss the importance of the a-priori valence of the smells in post-choice preference modulation.